AI statistics get source tracing

A statistic can sound solid because an AI assistant states it confidently and attaches a citation. But a citation may show where the assistant found a number, not where the number came from.

Trakkr Research asked five AI assistants 200 questions that called for a number, then traced a random sample of 1,285 statistics through their citations. Four in ten trails reached a study that supported the statistic. Another 8% reached a study, but the number or its context had changed. In 40% of cases, the researchers couldn’t check the source, and 11% of trails led nowhere. One example was the claim that 44% of salespeople give up after one follow-up. All five assistants gave that figure, but the trail ended at pages that cited no research. The investigation drew 98 Hacker News points and 49 comments, a sign that checking evidence matters when AI gives a fluent answer. Read the original report.

For a business owner, the point isn’t to distrust every number an AI gives you. It’s to check the evidence before putting a statistic into a sales deck, report, or decision. A citation is a starting point, not proof. This is the kind of source-checking process to use before relying on an AI answer. Learn more.